Efficient Inverse Design of Topological Photonic Crystals for Nonlinear Harmonic Generation via Combined Neural Networks

Topological photonic crystals (TPCs) have been extensively utilized in developing practical applications such as enhancing nonlinear harmonic generation. However, it is still a challenge to design topological photonic crystals for nonlinear harmonic generation. In this paper, we propose an inverse design method of a topological photonic crystal for nonlinear harmonic generation. We numerically calculated the band structure of a two-dimensional valley photonic crystal with a diamond lattice and generate training data. Subsequently, a combined network model was constructed and trained; it realized the inverse design from band structures to structural parameters of the photonic crystals and effectively solved the one-to-many problem brought about by a single inverse network model. On this basis, our target is to achieve on-demand design of the energy bands of valley photonic crystals and to utilize the structural parameters of the inverse design for nonlinear harmonic generation. The method demonstrated the effectiveness and flexibility of combined networks in the inverse design of topological photonic crystals, which can be generalized to other optimization problems of topological photonic crystals.